Re: ACTION A-641-13

> 
> This section says "the supplied value must be of type xs:string or xs:untypedAtomic" (it doesn't say it can be a derived type of xs:string or xs:untypedAtomic). Neither is true (it is xs:NMTOKEN), so I think the cast should fail.

If a value is an instance of xs:NMTOKEN then it is also an instance of xs:string.
> 
> Conversely, if the expression was " 'a' castable as xs:NMTOKENS", the cast would succeed.
> 
> This seems to be expressed in 18.3 in the Note with "This approach does not work with list types, because it is not permitted to use a value of type xs:NMTOKEN* as input to the constructor function xs:NMTOKENS"

That sentence is not quite accurate. Firstly, "not permitted" is loose language; everything is permitted, including things like division by zero that always result in an error. Secondly, in the special case of an xs:NMTOKEN* instance of length 1, the operation will actually succeed.

> 
> I think the same is true for the other list types, xs:ENTITIES(x) is never castable to xs:ENTITIES, and xs:IDREFS(x) is never castable to xs:IDREFS.

It's generally true that if L is a list type whose item type T is a subtype of xs:string, and whose length facets permit a list of length one, then casting an instance of T to L will succeed; and since an instance of T is also an instance of T*, it follows that in this particular case, the cast is idempotent.

Michael Kay
Saxonica